{% autoescape true %} iKnow - NTU AI Final Team9


Reveal Your Mind


Gaming is a fundamental field in artificial intelligence.
We think mind-reading is an interesting and challenging kind of games.
However, we find there are some drawbacks with naïve models that can solve the designing problem.
For example, users need to answer all questions if we use SVM, and decision tree cannot tolerate wrong answers.



Our dataset includes singers from 1998 to 2015 from hitFM. Singers features are crawled from wiki, KKBOX, TaipeiArena, GMA, Mojim

Difference Measurement

We use L2-norm distance as a heuristic to compute the differences between all singers and user’s answers so far. Singers having minimum and second smallest distance are kept as candidates.

Score criteria

Both reward and punishment are used. For each question, we assigned 5 points to whom satisfying the answer and minus 2 points for the selected singer if the answer is 'no'.

Question Generation

Sample features and apply decision tree 100 times using entropy as criterion. Randomly choose a feature based on the frequency that the feature is selected as root’s branching factor.

{% endautoescape %}